Page 1 of 32
Mercedes E350T Emission System Analysis Felix Domke – 2020/09/28
Contents 1 Introduction, Expertise ......................................................................................................................... 3
2 Abstract ................................................................................................................................................. 3
3 Problem Statement ............................................................................................................................... 4
4 Methodology ......................................................................................................................................... 4
4.1 Data Collection .............................................................................................................................. 4
4.2 Software Analysis .......................................................................................................................... 5
4.3 Behavioral Comparison ................................................................................................................. 5
5 Limitations of the described approach ................................................................................................. 5
6 General .................................................................................................................................................. 6
6.1 The Ammonia Load Model ............................................................................................................ 6
6.2 The Alternative Model .................................................................................................................. 7
7 Defeat Devices ...................................................................................................................................... 8
7.1 Test Cycle Behavior ....................................................................................................................... 8
7.2 Physical Limitations vs. Parameterization of the defeat devices .................................................. 9
8 Defeat Device #1: Exhaust Gas Mass Flow Limit ................................................................................. 10
8.1 Background ................................................................................................................................. 10
8.2 Summary ..................................................................................................................................... 10
8.3 Details ......................................................................................................................................... 10
8.4 Behavior in Alternative Model .................................................................................................... 13
8.5 Changes via Software Update ..................................................................................................... 14
8.6 Limitation of the Analysis ............................................................................................................ 15
9 Defeat Device #2: NOx Mass Flow ....................................................................................................... 16
9.1 Background ................................................................................................................................. 16
9.2 Details ......................................................................................................................................... 16
9.3 Changes via Software Update ..................................................................................................... 18
9.4 Limitation of the Analysis ............................................................................................................ 18
10 Defeat Device #3: Intake Air Temperature ..................................................................................... 19
10.1 Background ................................................................................................................................. 19
10.2 Changes via Software Update ..................................................................................................... 19
Page 2 of 32
11 Defeat Device #4: Restart Protection ............................................................................................. 20
11.1 Background ................................................................................................................................. 20
11.2 Details ......................................................................................................................................... 20
11.3 Changes via Software Update ..................................................................................................... 20
12 Defeat Device #5: SCR Temperature ............................................................................................... 21
12.1 Background ................................................................................................................................. 21
12.2 Details ......................................................................................................................................... 22
12.3 Changes to calibration data in software update ......................................................................... 24
12.4 Summary ..................................................................................................................................... 24
12.5 Limitations of the Analysis .......................................................................................................... 24
13 Defeat Device #6: AdBlue average consumption ............................................................................ 25
13.1 Background ................................................................................................................................. 25
13.2 Details ......................................................................................................................................... 25
13.3 Changes via Software Update ..................................................................................................... 25
13.4 Limitations of the Analysis .......................................................................................................... 25
14 Defeat Device #7 (EGR): Engine Start temperature ........................................................................ 26
14.1 Background ................................................................................................................................. 26
14.2 Details ......................................................................................................................................... 26
14.3 Correlation to the NEDC test cycle ............................................................................................. 26
14.4 Changes via Software Update ..................................................................................................... 26
15 Defeat Device #8 (EGR): “Hot & Idle” ............................................................................................. 27
15.1 Changes via Software Update ..................................................................................................... 28
16 Improvements for the updated version .......................................................................................... 29
16.1 Limitations of the analysis .......................................................................................................... 29
16.2 Details ......................................................................................................................................... 29
16.3 Effect on AdBlue consumption ................................................................................................... 30
16.4 Limitations on the analysis .......................................................................................................... 30
16.4.1 Summary ............................................................................................................................. 30
16.4.2 SCR: ..................................................................................................................................... 30
16.4.3 EGR: ..................................................................................................................................... 31
17 Conclusions ..................................................................................................................................... 31
Page 3 of 32
1 Introduction, Expertise I, Felix Domke, have been instructed to provide a report based on my expertise on the questions
detailed below.
My professional experience can be summarized as follows:
- Dipl.-Ing. FH (Master’s equivalent) in electronic engineering, Lübeck University of Applied
Sciences, 2008
- 20 years of experience in professional development of software for embedded devices, with
focus on security, security analysis and reliability
- 15 years of experience as an independent security researcher
- 5 years of experience in independent analysis of emissions-related functionality in Diesel
electronic control units, including publication of peer-reviewed paper on this topic1 and
presentations2,3
- Being active as an expert for the German Federal Motor Transport Authority to independently
analyze software implementation in regards to defeat devices, being active as an expert and
testifying as part of the parliamentary commission of inquiry in regards to vehicle emissions in
2017.
2 Abstract Daimler AG is a multinational automotive corporation that is the manufacturer of cars sold under the
brand of Mercedes-Benz. As cars are being sold to European market, they need to comply with a range
of emission regulations.
Allegations have been made that a significant number of vehicles that have been successfully tested for
compliance to these emission regulating rules perform very differently in normal-world scenarios,
exceeding emissions by a factor of 10X.
I was instructed to provide an expert opinion on whether Mercedes cars intentionally de-rate their
emissions control system during real-world driving. To do so, measurement data has been collected on a
modern (2015, conforming to Euro 6 emission standard) Mercedes car, equipped with a Selective
Catalytic Reaction (SCR) catalyst to reduce NOx emissions. Additionally, the calibration data from the on-
board engine ECU has been analyzed. The measurement data that was collected from the on-board
engine ECU of the car during real-world driving was then correlated to the calibration data.
The result of this study is that a number of defeat devices have been identified that de-rate the
efficiency of the emission control system during regular driving. These defeat devices will be discussed in
detail.
1 M. Contag et al., "How They Did It: An Analysis of Emission Defeat Devices in Modern
Automobiles," 2017 IEEE Symposium on Security and Privacy (SP), San Jose, CA, 2017, pp. 231-250, doi: 10.1109/SP.2017.66. 2 Felix Domke, Daniel Lange, “The exhaust emissions scandal („Dieselgate“)”, The 32nd Chaos
Communication Congress, 2015, Hamburg 3 Felix Domke, “Software Defined Emissions”, The 33rd Chaos Communication Congress, 2016, Hamburg
Page 4 of 32
3 Problem Statement There have been allegations of the usage of defeat devices in Mercedes cars that correspond to the
EURO 5 and EURO 6 standards. These allegations have been backed, for example by measurements of
real-world driving emissions using PEMS (Portable Emission Measurement System).
While these measurements indeed show a significantly higher amount of NOx emissions than what has
been observed in standardized emissions testing (for example NEDC in a lab for homologation), they
cannot – by the nature of being external – explain the difference in behavior of the car.
This report uses a different approach to explain these observed differences in emission behavior by
obtaining measurements not from external sensors, but by capturing information directly from the ECU
(Electronical Control Unit) that controls the engine and emissions management.
This gives a different view on the data – instead of treating the engine, engine controlling software and
the emission control system as a black box, and only observing the inputs (such as ambient temperature
and vehicle speed) and outputs (emissions as measured by external sensors in the PEMS device), this
approach can show intermediate data that is used internally in the ECU to control and steer the engine
behavior and emissions.
Because such internal behavior is typically not documented by the car manufacturer, existing diagnostic
facilities, together with detailed information about the software, have been used to obtain and analyze
the data.
In view of the amount of work involved in this investigation, only one vehicle has been analysed.
However, the Kraftfahrt Bundesamt has identified defeat devices in a wide range of Mercedes vehicles.
It is my opinion within a reasonable degree of engineering certainty that other Mercedes vehicles with
comparable engines and techniques contain comparable defeat devices.
4 Methodology The methodology used in this report is a combination of data collection, behavioral analysis (with
referencing calibration data) and interpretation of the results.
Because of the complexity and the amount of present information channels from the ECU (an ECU
internally processes in the order of more than 10000 signals) an iterative process has been used. As a
first stage, broad data collection has been used to profile the car during regular operation. After
analyzing the bulk collected data, data sections have been identified that are related to the emissions
control systems. Data collection has then been fine-tuned to provide higher-resolution data.
4.1 Data Collection Data collection has been done over the OBD-2 port. This port, located in the front left of the driver
compartment, is a standardized interface that allows the diagnostic tester to interface with the car and
the on-board ECUs. Using ECU-specific commands (“UDS ReadMemoryByAddress”), data internal to the
ECU can be read and logged. A custom device has been built that continuously reads out such data and
logs it to a storage medium. The bandwidth of the OBD-2 interface is limited, so depending on how
much data is being read, a full set of data can be obtained every few minutes (if reading out all available
information), or every few seconds (when limiting data selection to emissions-related areas). Using an
iterative process, the analysis has resulted in a minimized set of memory locations that still convey all
necessary information for the interest of understanding the potential employment of defeat devices.
4.2 Software Analysis The software of the ECU (Bosch EDC17CP57) has been obtained by reading out the ECU using a third-
party tool (“FLEX”, developed and sold by Magic Motorsports). Using the firmware and existing
documentation of similar Bosch ECUs allowed to first understand the general structure of the emissions
control system, and allowed to derive the exact location of both calibration data constants as well as
measurement points within the software. This allows to use existing data and calibration analysis tools
on this software.
4.3 Behavioral Comparison Two software versions have been compared:
1. Original software: The software that was originally installed (“CR61-GDB2-212SA-642LS-
EU6OPS_4x4_3S_NAG2-ME10<-> (13.07.2015 08:30:25)”)
2. Updated software: An updated software (“CR61-GHB0-212SA-642LS-
EU6OPS_4x4_3S_NAG2_VarB-ME21”) that was installed to the ECU to analyze the behavior of
the 2020 software update.
Before installing the updated software, the original ECU was cloned in order to leave the
integrity of the original ECU intact. After verifying functional correctness of the cloned ECU, the
cloned ECU was then updated with the new software.
By comparing the behavior of the software during similar driving conditions, and correlating these
behavioral changes with changes observed in the software and parameters, the differences between
software versions can be shown.
5 Limitations of the described approach By the nature of the described approach, only sampled data of real-world driving situations is available.
All data has been captured from a single car. In general, no attempt has been made to correlate the
driving behavior to an existing test standard (like RDE), but it has been attempted to cover typically used
scenarios (inner-urban, extra-urban, highway) at different conditions. No specific testing on a chassis
dynamometer has been done.
A single car has been selected for the tests:
Car Type: Mercedes E 350 BlueTEC 4MATIC T
Engine: OM642, 190kW, 2987 ccm, 6 cylinders
Model Year: 2015 (initial registration: 1/2016)
The car has been verified to not have any pending faults indicated by the on-board diagnostics.
Page 5 of 32
Page 6 of 32
6 General The analyzed car uses multiple strategies to reduce the amount of NOx emissions. In this case, the
algorithms and parameters for EGR (Exhaust Gas Recirculation) and SCR (Selective Catalytic Reaction)
have been analyzed. Additional factors for NOx emissions exist, but EGR and SCR have been prioritized as
they are most relevant.
In the Selective Catalytic Reaction for reducing NOx emissions, one of the most important control
variables for SCR is the amount of DEF (Diesel Exhaust Fluid, also called by the brand-name AdBlue, or
more generic reduction agent) dosing. On the car in question, the Engine ECU, a Bosch EDC17CP57,
controls in real-time how much DEF (if any) is being added into the system.
Within the SCR catalyst, the DEF is converted to ammonia, and reacts with the (and Oxygen) NOx to
Nitrogen and Water.
The measure of effectiveness is the “SCR removal efficiency” or “conversion efficiency”, which is the
ratio of NOx that is removed from the exhaust gas.
Physical conditions place an upper boundary on how effective the SCR system can be. Mostly, SCR
performance is limited by the following phenomena:
- If the SCR temperature is too low, injected DEF does not convert to ammonia.
- If the SCR temperature is too high, ammonia will oxidize directly.
- If the exhaust mass flow is too high (i.e. the exhaust speed is too high), the exhaust gas will not
have sufficient time to react with the SCR catalyst.
- Similarly, if the NOx mass flow is too high, the SCR cannot reduce all of the NOx.
For more details, specific SCR literature should be consulted. This list is non-exhaustive but shows that
there are legitimate physical limitations of an SCR system.
The SCR system in the tested car is designed to operate in one of two modes:
6.1 The Ammonia Load Model An SCR system is most efficient when enough ammonia is present in the SCR catalyst. In conditions
where the temperature of the SCR catalyst is too low to allow thermolysis of AdBlue to ammonia
(around ~150 °C), it is not possible to increase the ammonia concentration by adding more AdBlue.
The SCR catalyst however has the ability to store excessive ammonia up to a certain point. This property
can be exploited by filling the catalyst with ammonia. This is done by dosing more AdBlue than what
would be stoichiometrically required to reduce the current NOx load. In that case, extra ammonia will be
left over which then remains in the SCR catalyst.
Because the SCR catalyst has an upper limit of ammonia storage, when the SCR catalyst is fully loaded,
excessive ammonia will leave the catalyst and produce ammonia slip, which needs to be avoided. In
order to not over-fill the SCR catalyst, but still keep the SCR catalyst sufficiently filled whenever possible,
the state of SCR catalyst (amount of ammonia, reaction rate etc.) needs to be known to the ECU
software. Because no sensors exist that can directly measure this state, modelling is used instead. The
ECU has been carefully tuned to predict the behavior of the SCR catalyst based on input variables. A NOx
Page 7 of 32
sensor at the output of the SCR catalyst is used to verify and fine-tune the model during regular
operation, producing the very accurate model required to dose the right amount of AdBlue.
By utilizing the ammonia load model, the SCR catalyst can be kept filled most of the time, resulting in
very efficient NOx reduction, often exceeding 95% (i.e. removing >95% of NOx), as required by the
emission standards.
As any other model, this model is only an approximation of the real physical behavior. Within known
limitations, the model can be shown to work very accurately. However, in more extreme cases, for
example during high operating temperatures, high exhaust mass flow, high NOx concentration, the
model is not sufficiently accurate anymore. Relying on the ammonia load model in this case would risk
over- or under-dosing of AdBlue, which would result in either ammonia slippage or insufficient NOx
removal.
The on-board diagnostic of a car is required to detect these conditions, and is required to illuminate the
“Check Engine” light. Depending on how bad the detected issue is, it can also prevent the car from
operating at all. For example, if instead of AdBlue, water is filled into the AdBlue tank, the on-board
diagnostics are required to sense this and prevent starting the engine eventually.
In the context of potential defeat devices, as long as the ammonia load model is active, and
programmed intentional deviation (for example due to a defeat device attempting to limit AdBlue
consumption) from the “ideal” dosing would result in such detected over- or under-dosing. While it
would be possible to also manipulate the monitor process, this would have a significant impact on the
overall operation. No observation of usage of defeat devices on this car in the ammonia load model
have been made.
6.2 The Alternative Model As described, exceeding certain physical limits can reduce the accuracy of the ammonia load model to
the point where over- or under-dosing is risked. To remedy this problem, Bosch (and many other ECU
manufacturers) implemented an alternative control strategy which is vastly simplified. This alternative
control strategy is called “Alternative Model” in this report, but other names that have been used
include ”Pre-Control”, “Online-Dosing”, “Feed-Forward” etc.
In this mode, no attempt is made to keep the SCR catalyst at high ammonia levels. Instead, AdBlue is
dosed at the amount that is needed to reduce the current load of NOx only. The calculation for the
amount of dosed AdBlue is done stoichiometrically based on the estimated NOx load, reduced to the
“expected efficiency”.
The “expected efficiency” is a factor (between 0 and 1) which indicates how much of the NOx can be
removed. In ideal conditions, this reaches a very high number (>0.95), but depending on operating
conditions, the ECU calculates a much lower “expected efficiency”. A central observation of the analyzed
car is that this “expected efficiency” – probably better described from now on as “target efficiency” – is
held at a relatively low value in many driving conditions.
In the alternative model, the ECU does not attempt to reduce more NOx than what is given by the target
efficiency, as the amount of AdBlue that is dosed in reduced accordingly.
Page 8 of 32
Internally, many factors are combined to produce the “target efficiency”. The calculation starts with a
factor of “1” (meaning that theoretically all NOx can be reduced), but then applies a number of
correction factors (all between 0 and 1) that are multiplied together. These correction factors are based
on evaluating a number of conditions that are expected to reduce the possible reduction rate.
The observation in this car is that additional factors are employed that cannot be described to be based
on physical limitations – i.e. the “target efficiency” is intentionally de-rated to a lower value depending
on inputs that should not directly affect the NOx reduction in the SCR catalyst, i.e. factors that do not
resemble physical limitations, but rather are based on policies. These reductions, which can be
described as defeat devices, will be described further.
7 Defeat Devices Different defeat devices are employed in the analyzed car. For SCR, these defeat devices have the
following properties in common:
1. They trigger on physical properties that are generally needed to be monitored for extreme
conditions, such as temperature, mass flow etc.
2. However, they trigger routinely in what can be considered as normal “real-world” driving
conditions.
3. They are designed to have an effect well after being “triggered”, for example by using a large
hysteresis and or employing a “restart protection” (see below).
4. They greatly reduce the internal efficiency estimation of the SCR system, which drastically
reduces AdBlue dosing, which in turn generates a much higher NOx output.
In total, six defeat devices were identified that relate to the SCR-system. Three of these defeat devices
are subject to an “aging factor” that significantly lowers the thresholds at which these defeat devices are
activated. For two of these defeat devices, this happens at an aging ratio of 1% (i.e. very early in the
lifetime of the vehicle) and for another at 20%.
In addition, two defeat devices were identified that relate to the vehicle’s EGR-system that are triggered
in circumstances where a physical justification is lacking. When triggered, they reduce the operation of
the EGR system significantly.
The updated software uses optimized thresholds that produce a much-improved NOx performance,
showing that the car hardware is indeed much more capable. In the updated software, the defeat
devices are either neutralized or removed in their entirety.
7.1 Test Cycle Behavior No specific testing on a chassis dynamometer was performed on the analyzed car. From observation of
real-driving patterns that mimic those which are tested within the NEDC, certain predictions about the
behavior of the car during a test cycle can be made.
For some of the defeat devices, such as the intake air temperature switch, it is clear that these are not
triggered during a test cycle because the limits are clearly specified in the testing regulations, or because
the aging factor that triggers their increased operation has likely not been reached at the moment of
Page 9 of 32
testing. For other defeat devices that depend on the dynamic behavior of the car, it is the expectation
that these are not triggered during the test cycle, based on the predictions from regular car usage.
7.2 Physical Limitations vs. Parameterization of the defeat devices It is noteworthy that some of the defeat devices are parameterized to enforce “switch” to a less efficient
emission mode at a specified threshold.
Page 10 of 32
8 Defeat Device #1: Exhaust Gas Mass Flow Limit
8.1 Background Exhaust mass flow is the measurement of the amount of exhaust gas (burned and unburned fuel and air)
over time. The unit used in this paper is kg/h. The amount of exhaust mass flow is correlated with the
speed of the exhaust gas through the catalyst, and as such directly related to the time that is available in
the SCR catalyst to cause the NOx reduction reaction to happen. The engine ECU needs to track the
exhaust mass flow to detect SCR efficiency loss due to insufficient reaction time, and prevent over-
dosing of ammonia. As such, exhaust mass flow is a legitimate physical quantity that needs to be
modelled (or measured) and taken into account when calculating the SCR efficiency.
The SCR catalyst should be constructed to have a sufficient size (i.e. reaction chamber) that SCR
efficiency can be maintained at a satisfying rate during regular car operation. If the SCR catalyst is
designed as too small, a reduction of the SCR efficiency at higher engine rpm can be expected.
8.2 Summary A defeat device activates the Alternative Model when the exhaust mass flow exceeds a pre-determined
level (which depends on SCR aging). In real-driving scenarios, this limit was observed to be exceeded
typically around 100 km/h. Once in the Alternative Model, a mechanism restricts the target efficiency to
less than 60% in the majority of operating conditions.
8.3 Details As explained above, when the exhaust mass flow exceeds what the SCR catalyst can handle, exhaust gas
can escape the SCR catalyst without having had a chance to be fully reduced. If unhandled, this would
lead to an over-estimation of the SCR efficiency, causing over-dosing of AdBlue/ammonia, and would
yield excessive ammonia in the catalyst which would yield to ammonia slip. It seems a valid strategy to
observe the mass flow, and reduce the estimation for the SCR efficiency if excessive mass flow is
detected. Further, it is reasonable to assert that the SCR efficiency cannot be estimated accurately
enough for the ammonia load model anymore in this case, which requires a switch to the alternative
model. Such mechanism in itself can be required even in state-of-the-art emission control systems and is
generally not seen as an illegal defeat device because it is the direct result of a physical limitation.
In the case of the tested car, the limit however was observed at 170 kg/h – that is, if the filtered (i.e.
smoothed-out) exhaust mass flow, measured in kg/h of exhaust gas, is exceeded, a switch to the
alternative model is enforced. Further, a strong hysteresis (of -80 kg/h) is applied so that a switch back
to the load model requires the mass flow to be below 90 kg/h, which is routinely exceeded even driving
at lower speeds around 60km/h.
The exact limit is a function of an “SCR aging factor”. The aging factor is a percentage, where 100%
corresponds to a new catalyst, that is gradually de-graded (down to eventually 0%) over time. For this,
the engine ECU tracks the temperatures that the SCR catalyst has been exposed to, and models the
effect of aging. The tested car has an internal aging factor of ca. 69%.
It is noteworthy that the switch from 200 kg/h to 170 kg/h happens at an SCR aging factor of 99%, i.e.
only very new SCR catalysts will use a limit of 200 kg/h.
Page 11 of 32
Figure 1 - Exhaust Mass Flow Limit as a function of SCR aging factor, extracted from calibration data of the original software. At an aging factor of 99% the threshold is lowered from 200 kg/h to 170 kg/h. The lower bar indicates the hysteresis limit, i.e. when the initial limit was exceeded once, the mass flow needs to be below 90kg/h (or 120 kg/h for new SCR catalysts). (The internally tracked aging factor is a function of the conditions that the SCR catalyst was exposed to; thereby no simple aging-factor-to-time relationship exists.)
Because the mass flow rate is filtered, idling the engine for a short amount of time will not directly
switch back.
Exhaust mass flow rate is strongly depending on driving conditions. To estimate the effect for real-world
driving, the diagram below shows typical (i.e. observed in regular driving scenarios, typically on flat
roads) exhaust mass flow rates depending on vehicle speed. This is an estimation only, as exhaust mass
flow rate will also strongly depend on requested torque due to acceleration, road grade and vehicle
mass. But it can be seen that when driving faster than ~100 km/h it is highly likely that the upper
threshold will be exceeded, forcing a switch to the alternative model. At that point, the lower threshold
will become active, making it difficult to switch back in normal driving.
0
50
100
150
200
250
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Exh
aust
Mas
s Fl
ow
Lim
it [
kg/h
]
Aging Factor
Exhaust Mass Flow LimitExhaust Mass Flow limit (initial)
Exhaust Mass Flow limit (hysteresis)
Page 12 of 32
Figure 2 – Measurement data of observed Mass Flow (filtered) in Real-World Driving as a function of vehicle speed; the red and green bars indicate the limits for the “Exhaust Gas Mass Flow Limit” defeat device (green – initial limit, red – limit after being enabled due to hysteresis). It can be seen that typically between 100 km/h and 120km/h the limit (170 kg/h) is exceeded. Once the limit is exceeded, the lower bar limit activates.
Page 13 of 32
Figure 3 – Measurement data with original software; Two events visible: at around 16:10 (left black arrow), the exhaust mass flow exceeds the limit of 170kg/h, causing the defeat device to activate, enforcing a switch to the Alternative Model. At around 16:17 (right black arrow), the defeat device deactivates, eventually allowing a switch back to the Load Model. (Other defeat devices, for example the NOx mass flow limit, activate at the same time.) Bit names refer to the bit positions in the “prectl2”-Mode word that ECU software uses to determine when to switch to the alternative mode.
8.4 Behavior in Alternative Model Once switched to the Alternative model, the estimated efficiency is reduced by a factor that is looked up
from a map. The inputs to the map are the exhaust gas mass flow and the SCR temperature. The
efficiency correction is multiplicative – i.e. if the estimated efficiency is 80% before, and the map defines
a factor of 60%, the resulting efficiency would be calculated as 60% of 80% = 48% etc. The table below
shows this map – it can be seen that in most operating conditions, the estimated efficiency peaks at
60%. This means that even if all other conditions are perfect, no attempt is made to remove more than
60% of the NOx emissions.
No Dosing Load Model
Alternative
Model Load Model
Page 14 of 32
Mass Flow [kg/h]
50 75 100 150 180 200 225 250 300 500 600 700
SCR
Te
mp
erat
ure
[°C
]
130 9% 9% 9% 8% 8% 8% 8% 7% 7% 6% 6% 5% 160 18% 18% 17% 17% 17% 16% 15% 15% 14% 14% 11% 8% 180 25% 24% 23% 23% 22% 21% 20% 19% 18% 17% 15% 11% 200 34% 33% 33% 32% 32% 31% 29% 29% 27% 24% 19% 14%
225 60% 60% 60% 60% 60% 60% 58% 45% 38% 33% 25% 19% 250 60% 60% 60% 60% 60% 60% 58% 55% 51% 41% 32% 25% 280 60% 60% 60% 60% 60% 60% 60% 60% 55% 45% 36% 28% 300 60% 60% 60% 60% 60% 60% 60% 60% 55% 45% 65% 75% 380 60% 60% 60% 60% 60% 60% 60% 60% 55% 45% 65% 75% 400 60% 60% 60% 60% 60% 60% 60% 60% 55% 45% 40% 35% 480 60% 60% 60% 60% 60% 60% 60% 60% 55% 45% 40% 35% 500 75% 75% 75% 75% 75% 75% 75% 75% 60% 45% 40% 35%
Figure 4 - Efficiency Correction map via SCR-Temperature and Mass Flow, extracted from calibration data of original software. Reading example: At a mass flow of 150 kg/h, and an SCR Catalyst temperature of 250°C, the efficiency estimation would be reduced to 60% of the original value (i.e. by factor of 0.6x).
While operating in the Alternative Model already reduces the efficiency of the SCR system by the nature
of not attempting to store excessive ammonia in the catalyst for use on demand, this map further
reduces the target efficiency to peak out at 60% (aside from very high load, high temperature driving
conditions where 75% is encoded). This means that once the Alternative Model is activated – only in
this mode does the map take any effect – the ECU never attempts to supply sufficient AdBlue to reduce
more than 60% of the NOx load (outside of the discussed special case of very high mass flow).
While this efficiency correction map is the one that most drastically reduces SCR performance while in
the alternative model, other efficiency corrections are applied on top of this map, and may further
reduce efficiency.
8.5 Changes via Software Update The mechanism was changed in the following aspects:
1. Exceeding a Mass Flow limit will not anymore switch to the alternative model.
2. The efficiency correction map – still used when the alternative model is selected by a different
criterion – was updated to the following table:
Mass Flow [kg/h]
25 50 100 150 175 200 225 300 400 500 600 800
SCR
Te
mp
erat
ure
[°
C]
100 5% 5% 5% 5% 5% 5% 5% 5% 5% 5% 5% 5%
150 20% 20% 20% 20% 20% 20% 20% 20% 10% 10% 10% 5%
175 45% 45% 45% 45% 45% 45% 45% 25% 15% 10% 10% 5%
200 50% 50% 50% 50% 50% 50% 50% 35% 20% 12% 10% 5%
225 60% 60% 60% 60% 60% 60% 50% 40% 37% 33% 25% 10%
250 75% 75% 75% 75% 75% 75% 70% 58% 55% 50% 40% 20%
275 85% 85% 85% 85% 85% 85% 85% 75% 63% 55% 50% 25%
Page 15 of 32
300 90% 90% 90% 90% 90% 90% 90% 85% 65% 60% 60% 30%
350 95% 95% 95% 95% 95% 95% 95% 95% 75% 60% 60% 30%
450 95% 95% 95% 95% 95% 95% 95% 90% 75% 60% 60% 30%
550 80% 80% 80% 80% 80% 80% 80% 80% 50% 40% 40% 20%
650 50% 50% 50% 50% 50% 50% 50% 40% 20% 10% 10% 10% Figure 5 - Efficiency Correction map via SCR-Temperature and Mass Flow, extracted from calibration data of updated software.
8.6 Limitation of the Analysis Exhaust mass flow is the sum of fuel and air that discharged from the engine. The amount of injected
fuel depends on the torque set point as well as any additional exhaust heat requests, and as such is
depending on the exact driving situation. A strong correlation to vehicle speed exists, but many other
factors – such as how quickly the car is accelerated, the properties of the road and also ambient
conditions – exist. The exhaust mass flow quantities that have been observed for this report are
samples, and may not generally reflect the situation on other cars, other driving situations or other
ambient conditions.
The effect of SCR catalyst aging could not be measured on the tested car, as the tested car was
purchased with the SCR catalyst aging factor already below the encoded switch value. As such, only the
lowered threshold was used by the original software. The threshold of the aging factor is set at 99%,
thereby indicating that the switch to the stricter mass flow limit happens very early in the lifetime of the
catalyst, but the analysis approach did not allow this to be quantified in time or mileage.
The aging factor is an internal representation and does not necessarily reflect the real age of the catalyst
(as measured in time or mileage), but is rather a representation for the amount of wear to the catalyst,
based on the environmental conditions that the SCR was exposed to in its lifetime.
Page 16 of 32
9 Defeat Device #2: NOx Mass Flow
9.1 Background The SCR catalyst has a given capacity for reducing NOx. Excessive NOx mass flow will at some point
inhibit the ability for the catalyst to fully reduce NOx. Similar to the discussion for exhaust mass flow, this
effect must be taken into account to avoid the risk of over-dosing AdBlue, yielding ammonia slip.
The amount of NOx that is produced “upstream”, i.e. before the SCR catalyst, is – under the absence of
other post-engine NOx reduction methods – directly related to the operating conditions of the engine.
EGR, injection timing, rail pressure and other factors are used to balance keeping the NOx emissions low
versus other factors such as reducing particulate matter, fuel efficiency and subjective drivability.
9.2 Details As it can be seen in the diagram below, the NOx generated by the engine strongly depends on driving
behavior, and usage of EGR (and other NOx reduction techniques) in the engine. Nevertheless, a
correlation between vehicle speed and NOx emissions exists, even though other factors (such as
acceleration profiles) largely affect NOx emissions as well. The next diagram shows observed NOx
emissions, measured in mg/s, relative to vehicle speed.
Figure 6 – Measurement data with original software: observed raw NOx emissions during Real-World driving; the initial threshold is marked in green, when exceeded the lower threshold (marked in red) will activate (hysteresis).
In the tested car, a defeat device was found that will enforce the alternative model when the NOx mass
flow exceeds a calculated threshold. The threshold – observed at 15mg/s is marked in green in the
diagram. A switch back to the load model is only allowed when the low limit of 6.5mg/s is undercut
(hysteresis).
The threshold is depending on the discussed “SCR aging factor”. Noteworthy again is that the threshold
is initially set to 25 mg/s, but then ramps down to 15 mg/s at an aging factor of 99%, i.e. after very little
computed “aging”.
Page 17 of 32
Figure 7 - NOx Mass Flow Limit as a function of SCR catalyst aging, extracted from calibration data of original software. The upper (blue) line describes the initial threshold to switch to the Alternative Model, the lower limit describes the threshold that allows switching back (hysteresis).
This means that on excessive NOx production by the engine, the SCR system is further limited in reducing
these emissions only to a lower percentage, producing even larger emissions.
The heavy use of hysteresis must be noted here – once the limit of 15mg/s is exceeded once, the very
low limit of 6.5mg/s must be achieved first to switch back, which in typical driving scenarios is often
hard to do. While vehicle speed is correlated to NOx mass flow, other conditions (such as acceleration
profile) significantly affect NOx emissions, which does not allow to describe this limit as a function of
typical vehicle speed well. The distribution of observed NOx mass flow quantities from the figure above
can be used to evaluate the occurrences of such switches.
Once the Alternative Model is enforced, the SCR efficiency is already limited due to the discussed
"Exhaust Mass Flow/SCR Temperature” correction map, which is active in the alternative Model
regardless of the switch criteria. In the Alternative Model, an additional correction factor is calculated
based on the filtered NOx mass flow. The curve is configured so that when exceeding 40mg/s of NOx, the
target efficiency is further reduced pretty drastically.
0
5
10
15
20
25
30
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
NO
x M
ass
Flo
w L
imit
[m
g/s]
Aging Factor
NOx Mass Flow LimitNOx Mass Flow limit (initial)
NOx Mass Flow limit (hysteresis)
Page 18 of 32
Figure 8 - SCR efficiency correction factor as a function of NOx mass flow, extracted from calibration data of original software.
9.3 Changes via Software Update
- The change to the alternative model due to NOx mass flow has been completely removed. NOx
mass flow is still taken into account for calculating the expected SCR efficiency (and thereby
controlling the AdBlue dosing), but the alternative model is not enforced anymore as the result
of exceeding NOx emissions.
9.4 Limitation of the Analysis The amount of NOx produced in the engine depends on the amount of fuel burned as well as the
prevailing conditions within the engine, for example the amount of exhaust gas recirculation (EGR) used,
the rail pressure and injection timing.
NOx mass flow quantities that have been observed for this report are samples, and may not generally
reflect the situation on other cars, other driving situations or other ambient conditions.
The effect of SCR catalyst aging could not be measured on the tested car, as the tested car was
purchased with the SCR catalyst aging factor already below the encoded switch value. As such, only the
lowered threshold was used by the original software.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 50 100 150 200 250 300
NOx Mass Flow [mg/s]
SCR Efficiency CorrectionCorrection Factor
Page 19 of 32
10 Defeat Device #3: Intake Air Temperature
10.1 Background The ambient or intake air temperature should not significantly affect SCR operation. Nevertheless, it is
used in the tested car as one criterion to switch to the alternative model. A temperature switch
compares the intake air temperature with 12°C, and enforces a switch to the alternative model when
the intake air temperature is below this threshold. A switch back to the load model is only allowed when
the temperature then raises above 15°C (hysteresis).
When switched to the alternative model, either due to the intake air temperature or any other criteria,
efficiency is further reduced.
10.2 Changes via Software Update The mechanism was changed in the following aspects:
1. The mechanism of switching to the alternative model based on intake air temperature was
removed completely.
Page 20 of 32
11 Defeat Device #4: Restart Protection
11.1 Background The heavy use of hysteresis means that during regular driving it is hard to make the SCR system re-
enable the load model once the alternative model has been triggered by each individual defeat device.
In scenarios where the car stops and re-starts often (for example Taxi drivers), the hysteresis would not
be effective as the hysteresis is reset when the engine stops. In such scenarios, the SCR system would
continuously switch back to the load model. In the analyzed software, this is prevented even in such
scenarios with a “restart protection”, as described below.
11.2 Details An additional defeat device is present, which will observe the SCR temperature during the first 20
seconds. If it exceeds 50°C at any point, then for the next 240s the usage of the alternative model is
enforced unless the total amount of NOx emissions reaches 1800mg.
In practice, this means that if the car is stopped and re-started, the next 4 minutes will be forced to the
alternative model.
The logic that implements this behavior is complex and can be configuration to involve additional
criteria. For example, the total amount of NOx emissions is tracked, and upon exceeding a pre-
determined value, falling back into the ammonia load model is allowed if a minimum engine runtime
time has been exceeded, which can depend on the ammonia stored in the SCR catalyst. Some of this
additional logic is configured (using the calibration data) to have no effect on the analyzed vehicle.
11.3 Changes via Software Update - This mechanism has been removed completely.
Page 21 of 32
12 Defeat Device #5: SCR Temperature
12.1 Background The exhaust gas heats up the SCR catalyst as it passes by. As such, the temperature of the SCR catalyst is
directly depending on the heat flow volume of the exhaust. At higher speeds, with higher exhaust
temperatures and higher mass flow rates, higher SCR temperatures can be expected. The SCR catalyst
warms up from ambient temperature after the engine start. The SCR catalyst is cooled by the ambient
air, so at colder ambient temperatures generally slightly lower SCR temperatures can be expected.
The diagram below shows typical observed SCR temperatures in normal driving conditions. After the
engine start, the SCR catalyst slowly warms up from the exhaust gas until it reaches operating
temperatures. Warmup can be accelerated by doing additional fuel injections to produce higher
temperature exhaust gas.
During DPF (Diesel Particulate Filter) regenerations, high temperatures are required. These are achieved
by heating up the exhaust gas with torque-neutral fuel injections. This also heats up the SCR catalyst
(which sits further downstream), so high temperatures (up to 550°C typically) are reached during DPF
regeneration. At such high temperatures, NOx reduction via SCR is not feasible because any injected
AdBlue would directly oxidize, eventually producing more NOx. DPF events though are relatively rare
(even few hundred kilometers, lasting a few minutes).
During normal driving, temperatures between 180°C and 350°C can be expected in the SCR catalyst.
Page 22 of 32
Figure 9 - SCR temperature during Real-World driving scenarios. High SCR temperatures are used in DPF regeneration events, low temperatures are after engine start (warmup). The green bar indicates the initial temperature threshold. If this temperature is exceeded, the alternative model is enforced until the red bar (hysteresis) is underrun.
12.2 Details The ECU calculates a threshold as a function of the previously described “SCR catalyst aging” factor, as seen in the next figure.
Warmup phase
Page 23 of 32
Figure 10- SCR temperature threshold as a function of Aging Factor, extracted from ECU calibration data. It can be seen that the SCR temperature threshold rapidly changes when the aging factor reaches a particular state (internally modelled as “80%”).
The tested car has an internal aging factor of ca. 69%. This means that in normal operation, after giving
the SCR sufficient time to warm up, at roughly 120 km/h the SCR catalyst temperature typically
approaches 300°C, at which the SCR temperature defeat device will force a switch to the alternative
model.
Figure 11 – Observed measurement data during an event where high SCR temperature (exceeding 300°C) enforces a switch to the Alternative Model. (Note that other Defeat Devices are activated at the same time, already enforcing the Alternative model). During the second half (starting around 14:47h) a DPF (Diesel Particulate Filter) regeneration event can be observed, yielding very high exhaust temperatures even though vhiecle speeds are moderate.
275
280
285
290
295
300
305
310
315
320
325
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
SCR
Tem
per
atu
re T
hre
sho
ld [
°C]
Aging Factor
SCR Temperature TresholdTemperature (initial)
Temperature (hysteresis)
Page 24 of 32
12.3 Changes to calibration data in software update 1. The threshold was changed from 320%/300°C to 275°C; there is no effect of SCR catalyst aging
anymore.
2. The hysteresis was changed from 20K to 15K, so the lower threshold was changed from 280°C to
265°C.
12.4 Summary Observing SCR temperature is critical for estimating the NOx reduction efficiency. SCR operation is
generally limited during temperatures that are above a threshold that is commonly exceeded during
regular moderate-to-high-speed driving. The threshold was even lowered for the software update,
indicating that there is a physical necessity of reducing SCR usage.
However, the updated software manages to keep the SCR efficiency between 70% and 90%, even
though usage of the Ammonia Load Model is not possible in this case.
Figure 12 – Measurement data during event where high SCR temperature causes a switch to the alternative model in the updated software. SCR state: Red – dosing disabled, Green – Load model, Yellow – Alternative model. SCR efficiency, while reduced from when operating in the load model, is still maintained between 60%-80% when the alternative model is active.
12.5 Limitations of the Analysis The SCR catalyst temperature is a result of the heat input from the exhaust gas into the catalyst. As
such, it is correlated to the exhaust mass flow and temperature history. One major factor of these
quantities is thereby vehicle speed, but many other factors exist as well. The displayed correlation
between vehicle speed and SCR catalyst temperature thus should be seen as exemplary only and is not
defining the SCR temperature. The SCR catalyst temperatures that have been observed for this report
are samples, and may not generally reflect the situation on other cars, other driving situations or other
ambient conditions.
The effect of SCR catalyst aging could not be measured on the tested car, as the tested car was
purchased with the SCR catalyst aging factor already below the encoded switch value. As such, only the
lowered threshold was used by the original software.
SCR State
SCR efficiency
Page 25 of 32
13 Defeat Device #6: AdBlue average consumption
13.1 Background The amount of AdBlue consumption is directly proportional to the amount of ammonia that is made
available for NOx reduction. As such, the AdBlue average consumption is the result of the control
mechanisms that regulate the dosing of AdBlue.
Excessive AdBlue dosing can be a symptom of a malfunction in the emission control system; it could
thereby be argued that excessive AdBlue dosing should be prevented in itself to avoid over-dosing
AdBlue during system malfunctions. However, monitoring the AdBlue usage for the purpose of detecting
a malfunction is not a viable approach as it is a very slow indicator.
As such, AdBlue average consumption should not have an effect on the SCR efficiency calculations.
13.2 Details The ECU tracks the average AdBlue consumption, and will enforce a switch to the alternative model
when the AdBlue average consumption exceeds 820ml/1000km.
Additionally, once switched to the alternative model, the SCR target efficiency is reduced based on the
average consumption (and the accumulated NOx mass for the current driving cycle). The dominant
factor is the average consumption – starting at ~800ml/1000km, the efficiency is reduced significantly,
especially at higher consumption. This inherently cripples AdBlue consumption, regardless of the
physical necessity of dosing larger amounts.
Accumulated NOx mass in this driving cycle
5.0g 6.0g 20.0g 30.0g 31.0g 35.0g
Ave
rage
Ad
Blu
e co
nsu
mp
tio
n
[mg/
10
00
km]
400 100.0% 100.0% 110.0% 110.0% 120.0% 120.0%
650 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%
800 95.0% 100.0% 100.0% 100.0% 95.0% 90.0%
1000 70.0% 95.0% 95.0% 95.0% 70.0% 65.0%
1200 40.0% 90.0% 90.0% 60.0% 40.0% 35.0%
1400 17.0% 85.0% 22.0% 22.0% 17.0% 13.0%
1600 8.0% 10.0% 10.0% 10.0% 8.0% 5.0%
3000 3.0% 5.0% 5.0% 5.0% 3.0% 1.0% Figure 13 - SCR efficiency correction based on AdBlue average consumption, extracted from calibration data of original software.
13.3 Changes via Software Update This mechanism has been removed completely.
13.4 Limitations of the Analysis Due to the abundance of defeat devices, it was surprisingly hard to drive the car in a way that sufficient
AdBlue was being dosed to trip the AdBlue average consumption; the car operated in the Alternative
Model in a large portion of time. As such, the physical effect of the AdBlue switch was only observed for
very limited time. The effect of the SCR efficiency correction however could be seen in the observed
data.
Page 26 of 32
14 Defeat Device #7 (EGR): Engine Start temperature
14.1 Background The engine temperature at engine start can be a useful quantity to control additional measures that will
enhance engine behavior during startup.
However, once the engine has been warmed up, the behavior of the engine and the emission control
system should no longer depend on the engine start temperature.
Similarly, the maximum engine temperature that was observed during a particular driving cycle should
have no inherent effect on engine behavior.
14.2 Details Another defeat device reduces the EGR rate depending on the maximum observed engine temperature
during this driving cycle and the engine start temperature. In many cases, both of these values do not
reflect any property of the current engine operation, but affect the EGR at all times.
This thermal window has been designed so that full EGR operation is only possible when the engine was
started between 18°C and 35°C, and the engine temperature has never exceeded 86°C.
Maximum engine temperature
0 °C 20 °C 40 °C 50 °C 60 °C 84 °C 86 °C 95 °C
Engi
ne
star
t te
mp
erat
ure
-50 °C 0.0% 0.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%
17 °C 0.0% 0.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%
18 °C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0%
35 °C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0%
36 °C 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0%
70 °C 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0%
85 °C 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0%
90 °C 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0% Figure 14 – Relative EGR reduction depending on maximum engine temperature observed in this driving cycle and the engine start temperature, extracted from calibration data of original software. Marked in red are the operating conditions that can be expected in the NEDC testing.
14.3 Correlation to the NEDC test cycle Noteworthy is that conditions are estimated to be true for the NEDC testing at all times, especially in the
repeated ECE-15 part that only uses low engine power, but not satisfied during many regular driving
cycles.
14.4 Changes via Software Update While the logic has been retained in the software update, the map has now all-zero, meaning that no
EGR reduction happens based on engine start temperature and maximum engine temperature.
Page 27 of 32
15 Defeat Device #8 (EGR): “Hot & Idle” A defeat device reduces the EGR when the engine is warmed up (>80...90°C) but idling. This scenario
often happens when driving with moderate to high speeds (such as on a highway), and then continue
through an urban environment. In this case, the EGR used for the later parts of the driving cycle would
be reduced.
Current engine temperature
-40 °C -30 °C 0 °C 20 °C 40 °C 55 °C 75 °C 80 °C 90 °C 100 °C 103 °C 110 °C
rpm
800 100.0% 100.0% 50.0% 4.7% 2.1% 0.0% 0.0% 0.0% 80.0% 80.0% 100.0% 100.0%
1000 100.0% 100.0% 50.0% 5.0% 1.9% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
1200 100.0% 100.0% 50.0% 5.0% 1.9% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
1400 100.0% 100.0% 50.0% 10.0% 3.7% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
1600 100.0% 100.0% 50.0% 10.0% 3.7% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
1800 100.0% 100.0% 50.0% 10.0% 3.7% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
2000 100.0% 100.0% 50.0% 15.0% 6.8% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
2200 100.0% 100.0% 60.0% 20.0% 10.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
2400 100.0% 100.0% 70.0% 20.0% 10.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
2800 100.0% 100.0% 70.0% 20.0% 10.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
3200 100.0% 100.0% 70.0% 20.0% 10.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
3600 100.0% 100.0% 70.0% 20.0% 10.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
Figure 15 – Relative Reduction of EGR depending on current engine temperature and engine speed, extracted from calibration data of original software. Green areas indicate pure temperature-depending reduction, likely to prevent clogging of EGR. Red areas indicate a defeat device designed to reduce EGR when the engine has been running to operating temperature, and then idles.
Page 28 of 32
15.1 Changes via Software Update The software update retains the logic itself, but the map has been significantly changed. The operating
range has been extended, so that EGR is not as much reduced in low (<35°C) temperatures, similar to
improvements on the high-temperature side.
Most importantly, the entries that cause the engine to reduce EGR when idling have been removed.
Current engine temperature
-30 °C 0 °C 20 °C 35 °C 55 °C 105 °C 107 °C 110 °C 121 °C 124 °C 132 °C 135 °C
rpm
800 100.0% 45.0% 4.7% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
1000 100.0% 45.0% 5.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%
1200 100.0% 45.0% 5.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0%
1400 100.0% 45.0% 10.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0%
1800 100.0% 45.0% 10.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0%
2000 100.0% 40.0% 15.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0%
2200 100.0% 35.0% 20.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0%
2400 100.0% 35.0% 20.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0%
3000 100.0% 35.0% 20.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0% 100.0%
3500 100.0% 35.0% 20.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0% 100.0%
4000 100.0% 35.0% 20.0% 0.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0% 100.0%
4500 100.0% 35.0% 20.0% 0.0% 0.0% 0.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%
Figure 16 – Relative Reduction of EGR depending on current engine temperature and engine speed, extracted from calibration data of updated software. Green shows areas of improvement; the entries that reduce EGR in idle after usage have been removed completely. (Note the change of the temperature scale.)
Page 29 of 32
16 Improvements for the updated version
16.1 Limitations of the analysis To analyze the difference between the original software from 2016 with the updated software from end
of 2019 a method was needed to switch between two software versions, but otherwise leave the
operating conditions unchanged. This was done by first cloning the original engine ECU with a third-
party tool onto a second ECU, and then applying the updated software to this ECU. This was done to
preserve the integrity of the original ECU for further investigations.
After the installation of the updated ECU, no on-board diagnostic errors were flagged.
16.2 Details The defeat devices present in the original software cripple SCR usage in regular operation. This can be
quantified by observing the distribution of the target SCR efficiency. This has been plotted below – the X
axis shows the observed efficiency ranges (0...1 for 0%...100%), the Y axis shows how often this
efficiency range has been observed over time (unit-less). Higher bars indicate that these efficiency
ranges have been observed more often. Compared with the updated software, low efficiency ranges
(causing higher NOx rates) are much more prevalent; there are almost no times where the engine
operated with an SCR efficiency of > 80%.
Figure 17 – Measurement data: Distribution of SCR efficiency with original software. Higher bars indicate higher relative probability of a specific SCR efficiency to occur in real-world driving.
The updated software shows a drastically different picture. While there are still occurrences of low SCR
efficiencies, most of the time the engine operates between 80% and the upper 90%-range.
Rel
ativ
e p
rob
abili
ty (
un
itle
ss)
SCR efficiency factor
Page 30 of 32
Figure 18 – Measurement data, distribution of SCR efficiency with updated software. Higher bars indicate higher relative probability of a specific SCR efficiency to occur in real-world driving.
16.3 Effect on AdBlue consumption The increased efficiency comes, however, with the cost of increased AdBlue consumption. The average
AdBlue consumption with the old software was determined as 0.75l/1000km, with the updated
software, much higher values of 1.61l/1000km (>2X) are observed.
16.4 Limitations on the analysis By the nature of doing data collection in typical usage, the driving cycles that have been observed with
the original software do not match these from the updated software completely. Although the same
driving style and pattern has been employed, no systematic testing of specific driving conditions has
been attempted. As such, care must be taken when comparing the pre-update and post-update values.
However, it is believed, based on the observation of the software changes, the observed effect on
AdBlue dosing, that these values are representative of long-time usage, and thus can be compared.
16.4.1 Summary In summary, the following changes to the defeat devices have been found in the updated software:
16.4.2 SCR: - SCR Temperature limit was removed, SCR aging dependency was removed, effective limit
changed from 300°C to 275°C.
- Exhaust mass flow limit was removed as switch condition completely, efficiency correction map
was improved to allow operation in alternative mode while maintaining high SCR efficiency
- NOx mass flow limit was removed as switch condition, removed efficiency correction in
alternative model based on NOx mass flow.
- Air temperature dependency was removed.
Rel
ativ
e p
rob
abili
ty (
un
itle
ss)
SCR efficiency factor
Page 31 of 32
- ”Restart-Protection” was removed.
- Both the switch as well as efficiency correction based on AdBlue consumption were removed.
16.4.3 EGR: - EGR reduction based on engine start and maximum temperature was removed.
- EGR reduction when engine is hot and idling has been removed; EGR is still limited at very high
and low temperatures, but no longer special-cases idling.
In summary, all defeat devices that have been identified have been removed in the updated software.
17 Conclusions The general observation in the test vehicle is that the target efficiency of the SCR and EGR systems are
intentionally de-rated based on factors that are not based on physical necessities but rather on the
manufacturer’s policies.
This research identified the use of 8 defeat devices, of which six are related to the SCR system and two
to the EGR system. Three of the SCR-related devices are dependent on an “aging factor”. This factor
causes two of these devices (numbers 1 and 2, exhaust gas mass flow and NOx mass flow) to be
activated at an aging ratio of 1%, i.e. very early in the lifetime of the vehicle, and a third (number 5, SCR
temperature) at 20%. The majority of the lifetime of a vehicle hence the SCR-related defeat devices are
active.
For the SCR system, the defeat devices have the following in common:
- They trigger on physical properties that are generally needed to be monitored for extreme
conditions, such as temperature, mass flow etc.
- However, they trigger routinely in what can be considered as normal, “real world” driving
conditions.
- They are designed to have an effect well after being “triggered”, for example by using a large
hysteresis and/or employing a “restart protection”.
- They greatly reduce the internal efficiency estimation of the SCR system, which drastically
reduces AdBlue dosing, which in turn generates a much higher NOx output.
In relation to the EGR system, a thermal window was identified, as well as a defeat device that reduces
the EGR when the engine has been warmed up but is idling.
Due to the abundance of defeat devices, the car operated in the Alternative Model during most of the
driving, employing a moderate driving style.
For some of the defeat devices, it is clear that they not triggered during NEDC test circumstance because
the limits are clearly specified in the testing regulations, or because the aging factor that triggers their
increased operation has likely not been reached at the moment of testing. For other defeat devices, it is
the expectation that these are not triggered based on the predictions from regular car usage.
The updated software uses optimized thresholds that produce a much-improved NOx performance, but
does not require any other hardware changes. Each of the defeat devices that have been identified has
been removed in the updated software, resulting in significant improvements of the system’s overall
performance.